Claude can do most of the paperwork in project management: it turns a brief into a plan, writes the weekly status update and spots the task that will make you miss the date. The setup you pick decides what Claude still knows next week: a Claude Project, Cowork, Claude Code with a plan file, a connector to Asana, Linear or Jira, or a shared doc like a Context Board doc that every agent keeps current.
Every plan name, price and menu path below comes from Anthropic’s own pages, read on November 13, 2026. The Claude Code section is a test we ran that day on a small website relaunch plan.
ContextBolt is ours, so weigh the bias. Context Board is the last setup in the table, and each section says who should stop at an earlier one.
- You plan alone, in chat: make a Claude Project with your brief and the project instructions further down. Free covers up to five projects, and each one keeps its own memory.
- You want the weekly update written for you: a scheduled task in Cowork runs on a cadence, even with your computer asleep. It needs a paid plan (Pro is $20 a month).
- Your project lives in a folder or repo: keep a PLAN.md file and let Claude Code read and edit it. In our test, a fresh session picked up the plan in 9 seconds with no memory at all.
- Your team already works in Asana, Linear or Jira: add that tool’s connector, and Claude reads and updates the tasks where your team sees them.
- You use more than one agent, or other people need to see it: put the plan on a Context Board doc. Claude, Claude Code and Codex write in the same page, each line signed. $12 a month after 3 free days.
Which Claude setup fits your project
The question that sorts these five is where the plan lives between chats. Claude’s answer is only as current as the place it reads from.
| Setup | Where the plan lives | Next week, Claude knows | Who else sees it | What you pay |
|---|---|---|---|---|
| Claude Project | Your files and instructions in one project | The project’s own memory and files | Shared on Team and Enterprise | Free for up to 5 projects |
| Cowork | A project with a folder and scheduled tasks | Its memory, when it runs in the cloud | Your account | Pro and up for scheduled tasks |
| Claude Code | A PLAN.md file in your folder | Whatever the file says | Anyone with the folder | Pro and up |
| Asana, Linear or Jira connector | Your team’s project tool | Today’s tasks, read on each ask | Your whole team | The tool’s seats, plus Claude |
| Context Board doc | A page every agent writes in | The page, with the last 100 changes | Anyone you share it with | $12 a month, plus your Claude plan |
Pro is $20 a month billed monthly, or $17 a month billed yearly ($200 up front), on Anthropic’s pricing page. Max is $100 and Ultra is $200. Our guide to Claude’s usage limits covers how far each plan goes in a week.
Set up a Claude Project for project management
A Claude Project is a workspace inside Claude with its own files, instructions and memory. It suits a person who plans in chat and wants Claude to start each chat already knowing the project.
- Click + New Project in the upper right corner of claude.ai, then give it a name and a description.
- Click the + button in the project to add the brief, the plan, the budget and your last meeting notes.
- Click Set project instructions, paste the template from the next section, and click Save instructions.
- Start every project chat inside the project, never in a loose chat.
Those steps and button names are from Anthropic’s projects help page. Free accounts can make up to five projects. On paid plans, when the files near the context window’s limit, Claude turns on RAG mode by itself (it searches the files instead of reading them all), which Anthropic says expands capacity by up to 10 times.
Anthropic’s projects page also says a new version of projects is in beta, rolling out in stages starting with Claude Code for some Pro, Max and Ultra subscribers, with chat and Cowork to follow.
Memory is the reason to use a project. Anthropic’s memory page says each project has its own memory, and inside it Claude uses the project’s memory with your memory from chats outside projects, but not memory from other projects. So two clients in two projects never mix.
Where a project stops: the plan is a set of files you uploaded, and memory is Claude’s summary of past chats. When a date moves, you re-upload the plan or tell Claude in a chat. Sharing needs a Team or Enterprise plan. On Pro, nobody else sees it. We cover the wider memory question in nine ways to give Claude long-term memory.
Project instructions for project management (template)
Project instructions tell Claude how to behave in every chat inside the project. These are the instructions we gave Claude Code in the test further down, reworded for a Claude Project. Copy them and change the names.
You are the project manager for [project name]. Owner: [your name].
Source of truth
- The file "Plan" in this project is the plan. Read it before you answer
anything about dates, owners or tasks.
- Meeting notes are the files whose names start with "Notes". The newest
one wins when they disagree.
Status updates
- Four parts: Done, In progress, Blocked or at risk, Next.
- Every line has an owner and a date. Under 150 words.
- Join the dots: if a late task blocks a later one, say which.
When something changes
- Write the changed lines of the plan for me to paste back into the file,
and one dated line for the Decisions log saying what changed and why.
- Never move a milestone or a launch date without asking me first.
Never invent a date, owner, price or task. If something is missing, ask.The last two rules carry the weight. In our test, the Claude Code session asked before moving a milestone because the instructions said to, and it flagged the missing prices instead of guessing them.
Cowork: status updates on a schedule
Cowork is the side of Claude that works across your files and tools on a goal. Anthropic is folding it into the main app: its September 16 post says Claude Cowork and chat are merging into one Claude, rolling out to Pro, Max and Ultra. Until it reaches your account, Cowork sits next to Chat in the message box.
Cowork has its own projects. In the left menu, choose Projects, then +, then Start from scratch, Import from project or Use an existing folder. A Cowork project holds instructions, scheduled tasks, context (a folder on your computer, a linked chat project or a web address) and memory, and Anthropic’s help page says it needs the latest Claude Desktop app.
Scheduled tasks are the project-management feature. From Anthropic’s scheduled tasks page:
- Click Scheduled in the left sidebar, then New task.
- Pick Create with Claude or Set up manually.
- Give it a name, a prompt and a frequency: hourly, daily, weekly, on weekdays or manually. Add a folder if the task needs your files.
- Click Save.
A Monday 8 AM task that reads the plan and drafts the status update is the obvious first one. Scheduled tasks run remotely, so they run even when your computer is asleep or the app is closed, but a task that needs files or apps on your computer runs only there. They need a paid plan.
Where Cowork stops: Anthropic’s memory page says Cowork sessions that run on your own computer don’t use memory. The task that reads your local folder starts fresh each time, so the plan has to be in a file it can read. The status update lands in your Claude account, and you still forward it to the team.
Claude Code: keep the plan in a file (we tested it)
Claude Code is Anthropic’s coding agent, and it comes with Pro and up. It starts each session with no memory of the last one. It reads the folder instead, and that habit makes it a steady project manager.
We built a small test on November 13, 2026: a folder for a made-up website relaunch with three files. CLAUDE.md held the instructions from the template above. PLAN.md held four milestones, eight tasks with owners and dates, and a Decisions log. notes/2026-11-12-call.md held a call where the copywriter said one page was half done and the owner said the payment account was still waiting on verification.
Session 1. A fresh Claude Code session (version 2.1.322, Sonnet 5.5) got one line: “Give me this week’s status update for the team.” In 9.4 seconds it read both files and returned Done, In progress, Blocked and Next, with owners and dates. It joined two facts we never connected for it: the unverified payment account put the real-card booking test, due Nov 24, at risk. It also asked when the final prices would come, and moved no dates.
Session 2. A second fresh session got: “Priya says the About page copy slips to Nov 18. Leo finished the Home page build yesterday. Update the plan.” In 13.9 seconds it edited the file:
-- [ ] About page copy (Priya), due Nov 16
+- [ ] About page copy (Priya), due Nov 18
-- [ ] Home and Services pages built (Leo), due Nov 20
+- [x] Home page built (Leo), done Nov 12
+- [ ] Services page built (Leo), due Nov 20
...
+- Nov 13: About page copy moved from Nov 16 to Nov 18 at Priya's request (draft was half done as of Nov 12 call).
+- Nov 13: Home page build marked done (Leo finished Nov 12); task split so Services page build remains due Nov 20.Then it asked whether to move the copy milestone from Nov 16 to Nov 18, because the instructions said never to move one without asking. Neither session had any memory of the other. The file carried everything.
That’s the pattern the Google results for this search keep landing on. Flow AI’s January post describes Claude Code running most of its project management against Linear, Notion and GitHub, and Chris Blattman’s Claude Code skills (/setup-project-management, /weekly-review) build a weekly review from a project folder. If you plan features before Claude Code builds them, our guide to Claude Code plan mode covers the planning step inside one session.
Where Claude Code stops: the plan lives in a folder on your computer. Priya and Leo don’t open a terminal, your client never sees PLAN.md, and Claude in the browser can’t read it. Each person who needs the plan needs the folder, or a copy you send them.
Connect Claude to Asana, Linear or Jira
If your team already runs the project in a project tool, keep it there and let Claude work in it. Claude’s connectors directory lists the main ones, each marked as verified by Anthropic:
- Asana: “search, create, update, and track tasks, projects, and goals,” on Asana’s connector page.
- Linear: manage issues, projects and team workflows.
- Jira and Confluence: the Atlassian connector searches, reads and updates them, plus Bitbucket and Loom.
- Notion and monday.com have connectors too.
To add one in Claude, open Customize, then Connectors, press + next to Connectors, click the tool and choose Connect. Anthropic’s pricing page lists connectors on every plan, Free included. On Team and Enterprise, an owner turns them on first.
This route keeps one source of truth your whole team already reads. Claude visits when you ask: “What’s overdue in the relaunch project?”, or “Move every task Priya owns that’s due this week to Monday”. Between chats, the tool is as current as the last person who updated it. If you’re choosing a tool rather than connecting one, our Asana alternatives guide prices nine, with the AI each plan includes.
Where Claude’s memory of a project runs out
Each setup above keeps the plan inside one tool. That works until a second agent or a second person joins.
Say Claude in the browser wrote the plan, Claude Code builds the site, and Codex fixes a bug on the side. The Claude Project’s files and memory live in your claude.ai account. PLAN.md lives in a folder on your computer. Codex sees neither. You end up as the person who copies status between them, and your client sees nothing until you write it up.
A doc board on Context Board is one page all of them write in. Context Board is a web app at board.contextbolt.com, made by ContextBolt, that your agents keep current over MCP. A doc board holds pages inside pages, with to-dos, toggles, tables and live widgets inside a page. Any agent you connect reads a page as Markdown and writes to it: at the end, after a given line, in place of a section, or as a new page inside another.
What you get that a single tool can’t give you:
- One plan, every agent. Claude, Claude Code, Codex and Cursor connect with the same address. ChatGPT does too, on plans that allow a custom connector.
- Every line signed. A write glows once in that agent’s color with its name. With agent colors turned on in Settings, each line keeps a mark in its agent’s color, and hovering shows who wrote it. Board activity lists every change with Undo.
- History you can restore. Page history keeps the last 100 changes for 30 days, each with who made it. Restore this version puts one back.
- Live numbers in the plan. A widget in a page stays live while the board is open, such as your open issues from Linear or Jira, read from their own servers.
- Shared with people. Share a doc by email, where viewing is free for them, or as a small website at a public link. Editing needs their own subscription.
Start with one doc board named after the project. Add the board in Claude as a custom connector: Customize, then Connectors, then +, then Add custom connector, with the address the board gives you, and Allow on the board’s page. Claude’s Free plan can add one custom connector. Then ask Claude: “Write the relaunch plan on the Relaunch board: milestones, then tasks as to-dos with owners and dates, then a Decisions log.” Open Claude Code in your project, connect it to the same board, and ask it to tick the tasks it finishes. Both edits land on one page, each with its agent’s name. The full steps for each app are in our connect-your-agent docs.
Where Context Board stops: it doesn’t send email or Slack updates, and it has no team plan or seats. It doesn’t replace Jira for a 20-person team. It suits a founder or a small crew running projects with their agents.
Price: free for 3 days, then $12 a month. That includes every kind of board, so the dashboard with your live numbers comes with it. The board’s own chat runs on the Claude Code or Codex plan you already pay for, so there are no AI credits to buy.
Five prompts for the weekly project jobs
These work in any of the five setups, once Claude can read the plan:
- Monday status: “Write this week’s status update from the plan and the newest notes: Done, In progress, Blocked or at risk, Next. Owners and dates on every line.”
- Risk check: “Which tasks are due in the next 10 days with no progress noted? For each, what does it block?”
- Notes to tasks: “Turn these call notes into tasks with owners and dates. List anything without an owner as a question for me.”
- Client update: “Write a five-line update for the client from this week’s status. Leave out internal names and anything still undecided.”
- Change log: “Priya’s copy moves to Nov 18. Update the plan, add a dated line to the Decisions log, and tell me which milestone it puts at risk.”
The fifth prompt is the one most setups get wrong when the instructions don’t say “ask before moving a milestone”. Put that rule in, whichever tool you use. For a launch, our product launch checklist lists what usually goes in the plan.
Start with this week’s status
Pick the setup from the table, paste the project instructions in, and ask for this week’s status update. If the answer names an owner and a date on every line, the setup works. If more than one agent or person needs that answer, start a Context Board doc for the plan and connect Claude to it.